Agents going idle = 0 productivity

[lucascastellanibr]

I really liked how the AO works, but agents keep going idle and having to manually click to post is killing productivity.
If we have an agent orchestrator it should resolve things alone, he is the Architect Engineer.
Any way to solve this?

[tamish560]

The idle state is by design, not a bug. AO uses a human-in-the-loop model where the agent pauses before taking consequential actions (commits, file writes, running commands that change state). This is a safety boundary, not a limitation.

If you want less manual clicking, there are a few things you can do:

1. Check your agent's config. Some agents (like Claude Code) have a --dangerously-skip-permissions flag or equivalent that lets the agent proceed without pausing. AO passes these through based on your spawn config.

2. The "waiting for response" state often means the agent is asking for confirmation on a specific action. You can respond inline from the AO terminal pane instead of clicking through the UI.

3. For CI-bound workflows, if the agent is waiting because a test failed, AO can be configured to auto-retry failed test runs. Check the agent config YAML for retry policies.

4. If the agent is genuinely stuck (not just paused for approval), it usually means it hit a context limit or the task was ambiguous. Narrowing the prompt helps a lot here.

The architecture is intentionally conservative because autonomous agents that can write code, commit, and push without any checkpoint are a security risk in production repos. The idle state is the tradeoff for that safety.